Introduction
Intersections represent the most critical conflict zones in urban transportation networks, accounting for over 80% of pedestrian-involved accidents. While traditional safety models emphasize macro-factors such as traffic volume, speed limits, and weather conditions, the micro-level geometry of the built environment—specifically how buildings obstruct a driver's field of view—has remained largely unquantified. This research addresses this gap by integrating architectural data into traffic risk modeling to understand how the "urban canyon" influences collision risk.
Methodology
The study introduces a systematic approach to quantify visibility through a "View Percentage Calculation." The process follows a structured pipeline from data collection to model evaluation.
- Spatial Buffering: Accidents were geocoded to intersections using a 33.3-meter radius to isolate intersection-related incidents.
- Field of View Modeling: A driver’s effective field of view was modeled as an 80-degree sector, aligned with established visibility standards.
- Visibility Algorithm: An algorithm calculated the "unobstructed view percentage" by identifying angles blocked by building polygons sourced from OpenStreetMap.
- Dynamic Interpolation: Multiple viewpoints were interpolated along road segments to simulate the changing visibility a driver experiences when approaching an intersection.
- Statistical Modeling: A Generalized Linear Model (GLM) with a Poisson distribution was employed to analyze accident frequencies, comparing a baseline model against one incorporating the visibility metric.
Data Sources
The research integrated four primary pillars of data to build and validate the predictive models.
The primary model was developed using data from 935 intersections in London. To ensure generalizability, the findings were validated against a separate dataset of 243 intersections in Manchester, confirming that the relationship between building geometry and risk holds across different urban layouts.
Results and Discussion
The inclusion of "view percentage" as a variable significantly enhanced the model's performance. In the London dataset, the Akaike Information Criterion (AIC) dropped by over 322 units, indicating that visibility is a primary determinant of intersection safety. The final model achieved a Pseudo R-squared of 0.9318, a notable improvement over the baseline (0.9046).
The Visibility Paradox
A key finding was the positive coefficient (1.7450) for the visibility variable. Contrary to intuition, as the unobstructed view percentage increased, the expected frequency of accidents also increased. This suggests a "visibility paradox" likely driven by risk compensation: drivers in wide-open environments may drive more aggressively or with less attention, whereas "blind" intersections may naturally compel drivers to exercise greater caution.
Furthermore, intersections on primary roads showed a significant increase in accident logs (0.615), confirming that high-volume road complexity compounds the risks associated with visibility restrictions.
Conclusion and Future Work
This study demonstrates that building geometry is a robust, statistically significant predictor of urban traffic risk. While the model provides a breakthrough in spatial safety analysis, it currently focuses on static building footprints and does not account for dynamic obstructions like parked cars or vegetation.
Future research should integrate granular "live" traffic data and explore spatial autocorrelation between neighboring intersections. Ultimately, these findings provide urban planners with a tool to evaluate the safety implications of new architectural developments before construction begins, moving toward a more proactive approach to urban safety.
Summary updated